Four workloads, one operations team

Every one of these runs on the same clusters and the same on-call practice. What changes is the software configuration and where the boundary of responsibility sits.

AI Training
Multi-node training that stays up for weeks
Dense GPU nodes with high-bandwidth interconnect, wired for distributed training runs that cannot afford to restart from scratch.
- RDMA fabric
- Multi-node
- Shared NVMe
- 24 / 7
AI Inference
Serving capacity behind an OpenAI-compatible endpoint
Production inference on vLLM, SGLang or TensorRT-LLM, metered per token and reachable through a single API surface.
- OpenAI-compatible
- vLLM / SGLang / TRT-LLM
- Per token
- Multi-node
Fine-tuning
From base model to a checkpoint you can actually serve
LoRA, QLoRA and full-parameter fine-tuning, followed by the quantization and evaluation work that makes the result deployable.
- LoRA / QLoRA / Full
- GGUF / FP8 / GPTQ / AWQ
- IQ1_S → Q8_0
- 22 repositories
On-premise
Your models, your hardware, your network boundary
Localized deployment for teams whose data cannot leave the building, including non-NVIDIA accelerators.
- 5 vendors
- Air-gap capable
- LLMFabric
- Apache 2.0